参考资料:
https://leetcode-cn.com/problems/minimum-window-substring/solution/zui-xiao-fu-gai-zi-chuan-by-leetcode-2/
直觉
本问题要求我们返回字符串SS 中包含字符串TT的全部字符的最小窗口。我们称包含TT的全部字母的窗口为可行窗口。
可以用简单的滑动窗口法来解决本问题。
在滑动窗口类型的问题中都会有两个指针。一个用于延伸现有窗口的 rightright指针,和一个用于收缩窗口的leftleft 指针。在任意时刻,只有一个指针运动,而另一个保持静止。
本题的解法很符合直觉。我们通过移动right指针不断扩张窗口。当窗口包含全部所需的字符后,如果能收缩,我们就收缩窗口直到得到最小窗口。
答案是最小的可行窗口。
举个例子,S = “ABAACBAB”,T = “ABC”。则问题答案是 “ACB” ,下图是可行窗口中的一个。
初始,leftleft指针和rightright指针都指向SS的第一个元素.
将 rightright 指针右移,扩张窗口,直到得到一个可行窗口,亦即包含TT的全部字母的窗口。
得到可行的窗口后,将lefttleftt指针逐个右移,若得到的窗口依然可行,则更新最小窗口大小。
若窗口不再可行,则跳转至 22。
重复以上步骤,直到遍历完全部窗口。返回最小的窗口。
class Solution {
public String minWindow(String s, String t) {
if (s.length() == 0 || t.length() == 0) {
return "";
}
// Dictionary which keeps a count of all the unique characters in t.
Map<Character, Integer> dictT = new HashMap<Character, Integer>();
for (int i = 0; i < t.length(); i++) {
int count = dictT.getOrDefault(t.charAt(i), 0);
dictT.put(t.charAt(i), count + 1);
}
// Number of unique characters in t, which need to be present in the desired window.
int required = dictT.size();
// Left and Right pointer
int l = 0, r = 0;
// formed is used to keep track of how many unique characters in t
// are present in the current window in its desired frequency.
// e.g. if t is "AABC" then the window must have two A's, one B and one C.
// Thus formed would be = 3 when all these conditions are met.
int formed = 0;
// Dictionary which keeps a count of all the unique characters in the current window.
Map<Character, Integer> windowCounts = new HashMap<Character, Integer>();
// ans list of the form (window length, left, right)
int[] ans = {-1, 0, 0};
while (r < s.length()) {
// Add one character from the right to the window
char c = s.charAt(r);
int count = windowCounts.getOrDefault(c, 0);
windowCounts.put(c, count + 1);
// If the frequency of the current character added equals to the
// desired count in t then increment the formed count by 1.
if (dictT.containsKey(c) && windowCounts.get(c).intValue() == dictT.get(c).intValue()) {
formed++;
}
// Try and co***act the window till the point where it ceases to be 'desirable'.
while (l <= r && formed == required) {
c = s.charAt(l);
// Save the smallest window until now.
if (ans[0] == -1 || r - l + 1 < ans[0]) {
ans[0] = r - l + 1;
ans[1] = l;
ans[2] = r;
}
// The character at the position pointed by the
// `Left` pointer is no longer a part of the window.
windowCounts.put(c, windowCounts.get(c) - 1);
if (dictT.containsKey(c) && windowCounts.get(c).intValue() < dictT.get(c).intValue()) {
formed--;
}
// Move the left pointer ahead, this would help to look for a new window.
l++;
}
// Keep expanding the window once we are done co***acting.
r++;
}
return ans[0] == -1 ? "" : s.substring(ans[1], ans[2] + 1);
}
}
时间复杂度: O(|S| + |T|)O(∣S∣+∣T∣),其中 |S|∣S∣ 和 |T|∣T∣ 代表字符串 SS 和 TT的长度。在最坏的情况下,可能会对SS 中的每个元素遍历两遍,左指针和右指针各一遍。
空间复杂度: O(|S| + |T|)O(∣S∣+∣T∣)。当窗口大小等于|S|∣S∣时为 SS。当 |T|∣T∣ 包括全部唯一字符时为 TT 。
直觉
对上一方法进行改进,可以将时间复杂度下降到 O(2*|filtered_S| + |S| + |T|)O(2∗∣filtered_S∣+∣S∣+∣T∣),其中 filtered_Sfiltered_S 是从SS中去除所有在TT中不存在的元素后,得到的字符串。
当 |filtered_S| <<< |S|∣filtered_S∣<<<∣S∣时,优化效果显著。这种情况可能是由于TT 的长度远远小于SS,因此SS 中包括大量TT中不存在的自负。
算法
我们建立一个 filtered_Sfiltered_S列表,其中包括 SS 中的全部字符以及它们在SS的下标,但这些字符必须在 TT中出现。
S = “ABCDDDDDDEEAFFBC” T = “ABC”
filtered_S = [(0, ‘A’), (1, ‘B’), (2, ‘C’), (11, ‘A’), (14, ‘B’), (15, ‘C’)]
此处的(0, ‘A’)表示字符’A’ 在字符串SS的下表为0。
现在我们可以在更短的字符串filtered_Sfiltered_S中使用滑动窗口法。
import javafx.util.Pair;
class Solution {
public String minWindow(String s, String t) {
if (s.length() == 0 || t.length() == 0) {
return "";
}
Map<Character, Integer> dictT = new HashMap<Character, Integer>();
for (int i = 0; i < t.length(); i++) {
int count = dictT.getOrDefault(t.charAt(i), 0);
dictT.put(t.charAt(i), count + 1);
}
int required = dictT.size();
// Filter all the characters from s into a new list along with their index.
// The filtering criteria is that the character should be present in t.
List<Pair<Integer, Character>> filteredS = new ArrayList<Pair<Integer, Character>>();
for (int i = 0; i < s.length(); i++) {
char c = s.charAt(i);
if (dictT.containsKey(c)) {
filteredS.add(new Pair<Integer, Character>(i, c));
}
}
int l = 0, r = 0, formed = 0;
Map<Character, Integer> windowCounts = new HashMap<Character, Integer>();
int[] ans = {-1, 0, 0};
// Look for the characters only in the filtered list instead of entire s.
// This helps to reduce our search.
// Hence, we follow the sliding window approach on as small list.
while (r < filteredS.size()) {
char c = filteredS.get(r).getValue();
int count = windowCounts.getOrDefault(c, 0);
windowCounts.put(c, count + 1);
if (dictT.containsKey(c) && windowCounts.get(c).intValue() == dictT.get(c).intValue()) {
formed++;
}
// Try and co***act the window till the point where it ceases to be 'desirable'.
while (l <= r && formed == required) {
c = filteredS.get(l).getValue();
// Save the smallest window until now.
int end = filteredS.get(r).getKey();
int start = filteredS.get(l).getKey();
if (ans[0] == -1 || end - start + 1 < ans[0]) {
ans[0] = end - start + 1;
ans[1] = start;
ans[2] = end;
}
windowCounts.put(c, windowCounts.get(c) - 1);
if (dictT.containsKey(c) && windowCounts.get(c).intValue() < dictT.get(c).intValue()) {
formed--;
}
l++;
}
r++;
}
return ans[0] == -1 ? "" : s.substring(ans[1], ans[2] + 1);
}
}
复杂度分析
时间复杂度 : O(|S| + |T|)O(∣S∣+∣T∣), 其中 |S|∣S∣ 和 |T|∣T∣ 分别代表字符串SS 和 TT的长度。 本方法时间复杂度与方法一相同,但当|filtered_S|∣filtered_S∣<<<|S|∣S∣时,复杂度会下降,因为此时迭代次数是 2*|filtered_S| + |S| + |T|2∗∣filtered_S∣+∣S∣+∣T∣。
空间复杂度 : O(|S| + |T|)O(∣S∣+∣T∣)。
两次代码大部分相同,就是多了一个List
。然后左右指针left,right变成指向List里面的元素。查找范围有字符串S到了List。